Tensorflowlite在android缓冲区大小错误

Question

我正在尝试构建一个图像分类器android应用程序。我用keras制作了我的模型。模型如下：

model.add(MobileNetV2(include_top=False, weights='imagenet',input_shape=(224, 224, 3)))
model.add(GlobalAveragePooling2D())
model.add(Dropout(0.5))
model.add(Dense(3, activation='softmax'))
model.layers[0].trainable = False     
model.compile(optimizer='adam',  loss='categorical_crossentropy', metrics=['accuracy'])
model.summary()

输出：

Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
mobilenetv2_1.00_224 (Functi (None, 7, 7, 1280)        2257984   
_________________________________________________________________
global_average_pooling2d_2 ( (None, 1280)              0         
_________________________________________________________________
dropout_2 (Dropout)          (None, 1280)              0         
_________________________________________________________________
dense_1 (Dense)              (None, 3)                 3843      
=================================================================
Total params: 2,261,827
Trainable params: 3,843
Non-trainable params: 2,257,984

训练后，我使用转换模型

model = tf.keras.models.load_model('model.h5')
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
open(f"myModel.tflite", "wb").write(tflite_model)

对于android，代码为：

make_prediction.setOnClickListener(View.OnClickListener {
var resized = Bitmap.createScaledBitmap(bitmap, 224, 224, true)
val model = MyModel.newInstance(this)
var tbuffer = TensorImage.fromBitmap(resized)
var byteBuffer = tbuffer.buffer
// Creates inputs for reference.
val inputFeature0 = TensorBuffer.createFixedSize(intArrayOf(1, 224, 224, 3), DataType.FLOAT32)
inputFeature0.loadBuffer(byteBuffer)
// Runs model inference and gets result.
val outputs = model.process(inputFeature0)
val outputFeature0 = outputs.outputFeature0AsTensorBuffer
var max = getMax(outputFeature0.floatArray)
text_view.setText(labels[max])
// Releases model resources if no longer used.
model.close()
})

但每当我尝试运行我的应用程序时，它就会关闭，并且我在logcat中收到了这个错误。

java.lang.IllegalArgumentException: The size of byte buffer and the shape do not match.

如果我把我的图像的输入形状从224改为300，并在300输入形状上训练我的模型，然后插入android，我会得到一个错误。

java.lang.IllegalArgumentException: Cannot convert between a TensorFlowLite buffer with 1080000 bytes and a Java Buffer with 150528 bytes

任何形式的帮助都将不胜感激。

Answer 1

像使用一样使用它

make_prediction.setOnClickListener(View.OnClickListener {
var resized = Bitmap.createScaledBitmap(bitmap, 224, 224, true)
val model = MyModel.newInstance(this)
var tImage = TensorImage(DataType.FLOAT32)
var tensorImage = tImage.load(resized)
var byteBuffer = tensorImage.buffer
// Creates inputs for reference.
//val inputFeature0 = TensorBuffer.createFixedSize(intArrayOf(1, 224, 224, 3), DataType.FLOAT32)
//inputFeature0.loadBuffer(byteBuffer)
// Runs model inference and gets result.
val outputs = model.process(byteBuffer)
val outputFeature0 = outputs.outputFeature0AsTensorBuffer
var max = getMax(outputFeature0.floatArray)
text_view.setText(labels[max])
// Releases model resources if no longer used.
model.close()
})

然后与调试器一起检查问题是否仍然存在或

val outputFeature0 = outputs.outputFeature0AsTensorBuffer导致另一个。

如果你需要更多帮助，请给我打电话